One hackish way would be to use the grid
package functionality. Grab the output of each plot/map and store it as a gTree
object and then try to arrange the new objects in a grid.
library(tmap)
library(cowplot) # for plot_grid() function - good to arrange multiple plots into a grid
library(grid)
library(gridGraphics)
data(World)
tm_shape(World) + tm_polygons()
g1 <- grid.grab()
tm_shape(World[World$continent == 'South America', ]) + tm_polygons()
g2 <- grid.grab()
tm_shape(World[World$name == 'Brazil', ]) + tm_polygons()
g3 <- grid.grab()
# Try to arrange the plots into a grid using cowplot::plot_grid().
# First bind the p2 and p3 as one plot;
# adjust distance between them by forcing a NULL plot in between.
p23 <- plot_grid(g2, NULL, g3, rel_widths = c(1, -0.7, 1), nrow = 1)
plot_grid(g1, p23, nrow = 2, scale = c(0.8, 1))
I could not figure it out how to make it respond to the align
argument though :/ But maybe this puts you in some exploring direction or others can edit/improve this answer.
# Save the plot
ggsave(filename = "tmap-arrange-grid-1.png",
width = 10, height = 6, units = "cm", dpi = 150)

Note that, initially I thought that I could explore with adding a NULL
object to tmap_arrange
like tmap_arrange(p1, NULL, p2, p3, nrow = 2)
, but unfortunately, it does not accept it.
Another approach, inspired from this related question could be something along these lines:
library(grid)
grid.newpage()
pushViewport(viewport(layout = grid.layout(nrow = 2, ncol = 2)))
print(p1, vp = viewport(layout.pos.row = 1, layout.pos.col = 1:2))
print(p2, vp = viewport(layout.pos.row = 2, layout.pos.col = 1))
print(p3, vp = viewport(layout.pos.row = 2, layout.pos.col = 2))

Again, here, I didn't have the time to explore with aligning the plots perfectly, but others might improve this answer.